Controlling wellbore equipment using a hybrid deep generative physics neural network

Abstract

A system includes equipment for at least one of formation of, stimulation of, or production from a wellbore, a processor, and a non-transitory memory device. The processor is communicatively coupled to the equipment. The non-transitory memory device contains instructions executable by the processor to cause the processor to perform operations comprising training a hybrid deep generative physics neural network (HDGPNN), iteratively computing a plurality of projected values for wellbore variables using the HDGPNN, comparing the projected values to measured values, adjusting the projected values using the HDGPNN until the projected values match the measured values within a convergence criteria to produce an output value for at least one controllable parameter, and controlling the equipment by applying the output value for the at least one controllable parameter.

What is claimed is: 
     
       1. A system comprising:
 a processor; and 
 a non-transitory memory device comprising instructions that are executable by the processor to cause the processor to perform operations comprising:
 training a hybrid deep generative physics neural network (HDGPNN) using measured data from a hydrocarbon reservoir penetrable by a wellbore and using projections from a deep generative physics neural network (DGPNN); 
 iteratively computing a plurality of projected values for wellbore variables using the HDGPNN; 
 comparing the projected values to measured values; 
 adjusting the projected values using the HDGPNN until the projected values match the measured values within a convergence criteria to produce an output value for at least one controllable parameter; and 
 controlling equipment by applying the output value for the at least one controllable parameter, the equipment for at least one of formation of, stimulation of, and production from the wellbore. 
 
 
     
     
       2. The system of  claim 1 , wherein the DGPNN includes an operation to use a cost function based on a first principles physics model including boundary conditions. 
     
     
       3. The system of  claim 2 , wherein the measured data and the projections include location-based temperature values for the hydrocarbon reservoir and the first principles physics model comprises a steady diffusion model. 
     
     
       4. The system of  claim 1 , wherein the operations further comprise training the DGPNN by iteratively computing a set of outputs for an input combination. 
     
     
       5. The system of  claim 1 , wherein the operations further comprise denoising the measured values using the DGPNN to hybridize the measured values and the projections using a loss function based on mean squared error. 
     
     
       6. The system of  claim 5 , wherein the equipment includes a drilling tool and the at least one controllable parameter comprises at least one of weight-on-bit, rate of penetration, or revolutions per minute of a drill bit, applied based on location or orientation of the drilling tool. 
     
     
       7. The system of  claim 1 , wherein the DGPNN includes an operation to perform a nonlinear transformation of a hybrid input. 
     
     
       8. A method comprising:
 training, by a processor, a hybrid deep generative physics neural network (HDGPNN) using measured data from a hydrocarbon reservoir penetrable by a wellbore and projections from a deep generative physics neural network (DGPNN); 
 iteratively computing, by the processor, a plurality of projected values for wellbore variables using the HDGPNN; 
 comparing, by the processor, the projected values to measured values; 
 adjusting, by the processor, the projected values using the HDGPNN until the projected values match the measured values within a convergence criteria to produce an output value for at least one controllable parameter; and 
 controlling equipment by the processor applying the output value for the at least one controllable parameter, the equipment for at least one of formation of, stimulation of, and production from the wellbore. 
 
     
     
       9. The method of  claim 8 , wherein the DGPNN uses a cost function based on a first principles physics model including boundary conditions. 
     
     
       10. The method of  claim 9 , wherein the measured data and the projections include location-based temperature values for the hydrocarbon reservoir and the first principles physics model comprises a steady diffusion model. 
     
     
       11. The method of  claim 8 , further comprising denoising the measured values using the DGPNN to hybridize the measured values and the projections using a loss function based on mean squared error. 
     
     
       12. The method of  claim 11 , wherein the equipment includes a drilling tool and the at least one controllable parameter comprises at least one of weight-on-bit, rate of penetration, or revolutions per minute of a drill bit, applied based on location or orientation of the drilling tool. 
     
     
       13. The method of  claim 8 , wherein the DGPNN performs a nonlinear transformation of a hybrid input. 
     
     
       14. A non-transitory computer-readable medium that includes instructions that are executable by a processor for causing the processor to perform operations related to controlling equipment for at least one of formation of, stimulation of, and production from a wellbore, the operations comprising:
 training a hybrid deep generative physics neural network (HDGPNN) using measured data from a hydrocarbon reservoir penetrable by the wellbore and projections from a deep generative physics neural network (DGPNN); 
 iteratively computing a plurality of projected values for wellbore variables using the HDGPNN; 
 comparing the projected values to measured values; 
 adjusting the projected values using the HDGPNN until the projected values match the measured values within a convergence criteria to produce an output value for at least one controllable parameter; and 
 controlling the equipment by applying the output value for the at least one controllable parameter. 
 
     
     
       15. The non-transitory computer-readable medium of  claim 14 , wherein the DGPNN includes an operation to use a cost function based on a first principles physics model including boundary conditions. 
     
     
       16. The non-transitory computer-readable medium of  claim 15 , wherein the measured data and the projections include location-based temperature values for the hydrocarbon reservoir and the first principles physics model comprises a steady diffusion model. 
     
     
       17. The non-transitory computer-readable medium of  claim 14 , wherein the operations further comprise training the DGPNN by iteratively computing a set of outputs for an input combination. 
     
     
       18. The non-transitory computer-readable medium of  claim 14 , wherein the operations further comprise denoising the measured values using the DGPNN to hybridize the measured values and the projections using a loss function based on mean squared error. 
     
     
       19. The non-transitory computer-readable medium of  claim 18 , wherein the equipment includes a drilling tool and the at least one controllable parameter comprises at least one of weight-on-bit, rate of penetration, or revolutions per minute of a drill bit, applied based on location or orientation of the drilling tool. 
     
     
       20. The non-transitory computer-readable medium of  claim 14 , wherein the DGPNN includes an operation to perform a nonlinear transformations of a hybrid input.